<p>Lora trained in recreate Tianliang Duohe Fangdongye's artstyle. Just a little name...</p><p></p><h2 id="anima-preview-v2:-lokr">Anima Preview v2: LoKr</h2><p>I was able to train a lokr and it seems to be better for styles than regular loras (as they was on SD models). I tested them on Comfy<br /><br />Weight: .8 - 1</p><h2 id="anima-preview-v1:-lora">Anima Preview v1: LoRa</h2><p>I was playing with this model since their release. It's a pretty cool improvement for those models that requires a lot of resources for run. It feels like a illustrious v0.1 it needs several tricks to make it work, but using any lora, it gets really easy to use. <br />This was a test of train a Lora. I selected this one due the native style mess the hands rlly hard, and It works pretty well without any special config.</p><p>Try use it in a tag + natural language mix. Isn't consistent like is z-image but it is easier to use than Neta, more powerful than SDXL, uses almost the same vram than SDXL (lower if you use quantized model) . Is Slower than SDXL but it doesn't burn your computer like Flux or Auraflow does. <br /><br />Weight: 1 - 1.2</p><h2 id="pony-v1:">Pony v1:</h2><p>This one is maybe a little overtrained, try use a lower weight than 1 if you have problems.</p><p>Weight: .8 - .9</p>

Files with SHA256: 158e4a91902acef16f3b8a1068eed851a16eb564b5ef6246e131698c1c943644